Flat Fields for the Echelle Spectrograph

Calibration Lamp

A quartz calibration lamp is available on the instrument rotator, in the same optical train as the wavelength calibration lamps. This is the only flat field source which is sufficiently bright for use with this spectrograph; the dome flat is too faint. I used to take daytime sky flats, but I never discovered any use for them.

Manually operated neutral density filters can be inserted in front of the lamp: they cut the light by a factor of 0.3 or 0.1.

WARNING!

There is a comparison lens in the spectrograph which is driven over the slithead to simulate the telescope beam during comparison exposures. This is manually operated. There is no point in using it unless you are using the full slit length, with a mirror replacing the cross-dispersing grating. It is important to make sure that the comparison lens is out of the beam when you are actually observing... there seem to be little elves around the place who are always putting it in.

Strategy

The Problem

It seems natural to take flat field exposures through the same decker that will be used for observing, so that at any place on the CCD the flat field is measured at the same wavelength as the program objects. Since it is not possible to remotely switch between the flat field and arc lamps, it is normal to have the arc lamps selected during the night, and to use flat fields taken at a single telescope position during the afternoon.

Unfortunately, the spectrograph has enough flexure (1-2 pixels) in the cross-dispersion direction that the ends of the sky spectrum will sometimes shift off into the gaps between orders in the flat-field spectrum. Then you cannot flat-field all of the sky. This is OK if you are using a wide decker and can afford to throw away some sky pixels, but especially for faint objects and wide wavelength coverage, it rarely turns out that there is extra sky to be spared. What to do?

Solution #1: Using deckers of various lengths.

One approach is to take quartz lamp exposures through the decker which will be used for observing, and then also through the next larger decker. But for many setups the spectrum taken through the larger decker will have overlapping orders at the blue end. Maybe you can sort it out somehow...

Solution #2 (works only in blue): Not using a decker at all.

In the blue, where the CCD doesn't fringe, the pixel-to-pixel flat field characteristics have very little wavelength dependence. It is then possible to take the flat field spectra using the "open" decker position, letting lots of different wavelengths overlap at any place on the CCD. This spectrum can be reduced to produce a flat-field frame which covers the full CCD.

This works out to about 6000A with the Tek2K CCD on the long cameras, and ought to work with the Tek1K + folded schmidt. It may also work with the Loral 3K + Air Schmidt...I've never tried it there.

To reduce this type of flat field frame, try the following IRAF recipe:

use qtz lamp spectra taken with the decker wide open.
take the usual biases (16 minimum) and combine together with IMRED.CCDRED.ZEROCOMBINE
bias subtract and avsigclip together using IMRED.CCDRED.FLATCOMBINE
remove general overall shape of averaged flat field image using IMAGES.IMSURFIT, with parameter type_output="response". Rather high order chebyshev polynomials are usually required (for the Tek2048, try chebyshev, xord=50, yord=20, type_output=response).
use IMAGES.MEDIAN with ywindow=1 and xwindow > 2×(smallest surface feature which you wish to flat field out). For the IMSURFIT parameters suggested above, try using MEDIAN with xwindow=121). Divide median-filtered image into un-median-filtered image.
this should leave you with a beautiful flat field normalized to 1.0 except for CCD defects, dust features from dewar window, etc. If there are "shadows" from bad spots, it is usually because the median filter was working on a sloping background, so you have to try harder with imsurfit.
spots in this flat which go close to zero could be set to fairly large values at this time, in order to avoid wild points in data frames after dividing by the flat. PROTO.IMREPLACE can do this.

Number of exposures needed.

It depends on the range in intensity levels in a single exposure, and on the s:n you want. I usually take a short exposure to find out the intensity levels, and then calculate an exposure time that avoids saturating any spot on the CCD, and will give at least 300-500 counts at the faintest point of interest on the CCD. Note that you don't care about the flat field at spots which are so far outside the free spectral range that you will not get any counts from the object. You want to get a minimum of three exposures, in order to filter out cosmic rays. I generally take five; the whole flat-fielding process takes about half an hour.

J.Baldwin
28 Dec 1995